AIRIS — Aerial Intelligence for Reconstruction and Avian Surveillance

Single-Pass Drone Video to Accurate 3D Model Generation System (SIH Problem Statement 26158)

Live Command Center · Presentation · Demo · Setup · Requirements checklist

Actual source-textured reconstruction. Georeferenced to WGS 84 / UTM 32N with 518,786 point observations.

1. Project Information

2. Problem Statement

Conventional 3D photogrammetric mapping requires multiple drone passes, extensive overlap grids (≥80%), and lengthy ground station processing. In critical disaster response, rapid infrastructure inspection, and defense reconnaissance, only one continuous single-pass video is available.

Problem Statement 26158 challenges solutions to generate metric-accurate, georeferenced 3D terrain and structural models from a moving UAV's single-pass video, GPS coordinates, and flight metadata, exporting standardized GIS point clouds and textured 3D meshes.

• Spatial Accuracy: Target ≤1 m.
• Speed: Target <15 minutes processing for a 10-minute video.
• Visible-Scene Coverage: Dense extraction of visible facades, terrain, and structural boundaries.

3. AIRIS Architecture & Solution

AIRIS integrates pretrained multi-view geometry, visual camera pose graph optimization, and sensor telemetry fusion (GNSS + Barometric Altitude):

DRONE VIDEO
  → FRAME EXTRACTION
  → CAMERA / FLIGHT TRAJECTORY
  → AI RECONSTRUCTION
  → GEOREFERENCING (WGS 84 / UTM 32N)
  → 3D POINT CLOUD (518K points)
  → 3D MESH (646K triangles)
  → TEXTURE (Source Photographic Reprojection)
  → TERRAIN / STRUCTURE ANALYSIS
  → MEASUREMENT & INTELLIGENCE
  → EXPORT (GLB, PLY, LAS, GeoTIFF DSM, JSON)

The primary application is an interactive 3D Geospatial Command Center:

• Interactive 3D Viewport: Centerpiece WebGL visualizer with orbit, pan, zoom, and nadir orthographic camera views.
• Synchronized Telemetry Loop: Drone video, camera trajectory, 2D coverage map, and 3D camera viewpoints update synchronously along the 90-keyframe timeline.
• Georeferenced Precision: True UTM Zone 32N Easting, Northing, and altitude coordinates extracted from flight telemetry.
• Layer Visibility & Structural Classification: Toggles for Point Cloud, Mesh, Source Textures, Ground Grid, Wireframe, Terrain, Buildings, and Vegetation.
• Full-Screen Authentication Gateway: Clean dark aerospace console with OAuth and instant guest exploration.

4. Deliverables & Supported Products

• Point Cloud: .PLY (518,786 points, visual frame) and .LAS (UTM 32N georeferenced).
• 3D Polygonal Mesh: .GLB (source-textured surface) and .OBJ (Wavefront CAD format).
• Surface Model: .TIF (GeoTIFF Digital Surface Model - DSM).
• Evaluation & Diagnostics: .JSON (complete quantitative verification report).

5. Technology Stack

6. Installation & Verification

# Clone repository
git clone https://github.com/mouryas-aiml/AIRIS.git
cd AIRIS/viewer

# Install frontend dependencies
npm install

# Run automated verification test suite (72 unit and integration tests)
npm test

# Build production bundle
npm run build

7. License & Attribution

Developed for Smart India Hackathon (SIH 2026). Dataset references credit Zurich Urban MAV and affiliated open-source computer vision libraries.